Homepage | Ismail Harrando

Reseach

Natural Language Processing

My main interest is to extract high-level descriptors (document class, topics, entities..) from textual data.

Explainable zero-shot topic extraction using a common-sense knowledge graph

Harrando, I. Troncy, R.

3rd Conference on Language, Data and Knowledge (LDK'2021), Zaragoza, Spain, September 2021.

Wanna build a text classifier without any training data that can also explain its predictions? ZeSTE may be what you're looking for!

ProZe: Explainable and Prompt-Guided Zero-Shot Text Classification

Harrando, I. Reboud, A. Schleider, T. Ehrhart, T. Troncy, R.

IEEE Internet Computing, Special issue on knowledge-infused learning for computational social systems, July 2022.

To improve the state of the art on zero-shot text classification, we combine the explanatory power of a common-sense knowledge graph with the world knowledge contained in pretrained-language models by conditioning them on domain-specific prompts.

Apples to Apples: A Systematic Evaluation of Topic Models

Harrando, I. Troncy, R.

13th Conference on Recent Advances in NLP (RANLP'2021), Online, September 2021.

Topic Modeling Evaluation is an open problem in the Topic Modeling community. While the reliance on automatic evaluation remains more or less necessary to quickly assess the performance of a given topic model algorithm, there is no study that attempts to evaluate several algorithms in the literature given the same preprocessing, datasets, and metrics. That's what we did!

ToModAPI: A Topic Modeling API to Train, Use and Compare Topic Models

Lisena, P. Harrando, I. Troncy, R.

2nd Workshop for NLP Open Source Software (NLP-OSS @ EMNLP'2020)}, Online, November 2020.

This API is built to dynamically perform training, inference, and evaluation for different topic modeling techniques. The API grant common interfaces and command for accessing the different models, make easier to compare them.

Discovering interpretable topics by leveraging common sense knowledge

Harrando, I. Lisena, P. Troncy, R.

September 2021, 3rd Conference on Language, Data and Knowledge (LDK'2021)

How to make the results of topic modeling algorithms more understandable to humans? Try to add some common sense into the process :)

And cut! Exploring textual representations for media content segmentation and alignment

Harrando, I. Troncy, R.

2nd International Workshop on Data-driven Personalisation of Television (DataTV @ IMX'2021), Online, June 2021.

In this work, we present an approach to content segmentation that leverages topical coherence, language modeling and word embeddings to detect change of topics.

Named Entity Recognition as Graph Classification

Harrando, I. Troncy, R.

18th Extended Semantic Web Conference (ESWC'2021 - Poster Track), Online, June 2021.

Injecting real-world information (typically contained in Knowledge Graphs) and hand-crafted features into a pipeline for training end-to-end Natural Language Processing models is an open challenge. In this paper, we propose to approach the task of Named Entity Recognition, which is traditionally viewed as a Sequence Tagging problem, as a Graph Classification problem.

Multimodal Content Analysis

I study how to represent and extract information from multimedia content (e.g. videos) for narrative summarization, memorability prediction and recommendation.

Combining Semantic and Linguistic representations for Media Recommendation

Harrando, I. Troncy, R.

Multimedia Systems Journal, January 2023.